There is a collection of 5184x3156 images taken in the process of aerial photography, on which it is necessary to recognize steppe saigas. Examples of input data: №1 , №2 , №3 .

I encounter a similar task for the first time - I will be grateful if you advise ways to solve the problem (what area to dig for, how to approach). Also I will be glad to links to any ready-made software solutions to related problems.

    2 answers 2

    I cannot give any specific methods for solving the problem, but it is definitely necessary to use a two-dimensional image correlator. As algorithms, use the fast cross-correlation algorithm (based on FFT).

    Regards, maxspb89.

    • I did not quite understand the task, I thought that automatic real-time recognition was needed. ) - maxspb89

    Adobe Photoshop Extended helps to count objects in an image . Two options:

    • manual mode: click the mouse on each saiga found by the eyes;
    • automatic: count the number of selected areas. It is necessary to process the image with filters and level settings, etc. so that saigas are different from everything else, for example, in color (the background is white, they are black, but only they are black). It can be done. Then select by color (black), and automatically count the number of individual selected areas. With the homogeneity of the images, it is enough to set up such processing once, and you can analyze the images in batches.

    How I would process the picture:

    • I chose from the three channels the one in which the saikagi are most contrasting, and then I worked only with it;
    • using the method of frequency decomposition, he selected objects of saiga size into a separate layer;
    • levels, filter Find Edges or other achieved a white background without the "garbage" and black spots of saigas;
    • Select - Color range - select all black areas:
    • count them.

    With the same scale of pictures (and the size of saigas on them), you can drive the entire sequence of actions into File - Automate - Batch ... and apply to all images in the folder.